Survival Prediction of Cirrhosis Patients Using Polynomial Features and Differential Evolution

This research focuses on enhancing the accuracy of the Cirrhosis Patient Survival Prediction dataset from the UCI-MLR. Leveraging novel techniques such as feature engineering and hyperparameter tuning, the study employs a comprehensive dataset with attributes including drug administration, age, sex, bilirubin, cholesterol, and albumin levels, along with labels indicating the stage of liver cirrhosis. The chosen dataset, recognized for its reliability and widespread adoption in research, serves as a valuable benchmark for developing diagnostic models. The comparative analysis involves Gradient Boosting Classifier, Logistic Regression, K Nearest Neighbors Classifier, Decision Tree Classifier, Gaussian Naive Bayes Classifier, Linear Discriminant Analysis, and Support Vector Classifier. Further, hyperparameter tuning through Grid Search and Differential Evolution methods results in substantial accuracy improvements, particularly with the latter yielding a remarkable 9% increase with respect to the control dataset with no alterations. This work showcases the potential of advanced methodologies in optimizing cirrhosis prediction models, emphasizing the importance of feature engineering and hyperparameter tuning in achieving significant performance enhancements.